The
fuel
consumption
characteristics
of
heavy-duty
refrigerated
vehicles
are
studied
using
remote
monitoring
data,
so
as
to
explore
the
driving
travelling
on
actual
roads
and
their
laws.
First
all,
it
is
determined
describe
vehicle
level
by
100km
value,
calculate
analyse
monthly
quarterly
average
research
vehicle,
find
out
interval
change
rule
each
type
vehicle.
Then,
according
above
calculation
results,
a
random
forest
prediction
model
was
constructed
with
weight,
month
independent
variables.
Finally,
one
light,
medium
heavy
truck
used
verify
its
12-month
results
showed
that
effect
better
overall.
for
duty
reefer
trucks
in
this
paper
can
provide
support
government
regulate
vehicles.
Atmospheric Pollution Research,
Journal Year:
2024,
Volume and Issue:
15(7), P. 102152 - 102152
Published: April 22, 2024
PM2.5
has
caused
serious
harm
to
human
health
and
the
environment,
so
it
is
particularly
important
accurately
predict
concentration.
Aiming
at
problem
that
nonlinear
features
of
data
are
difficult
be
learned
accurately,
which
results
in
low
prediction
accuracy,
WOA-VMD-BiLSTM
hybrid
model
proposed
this
paper.
This
optimizes
parameters
variational
modal
decomposition
(VMD)
by
Whale
Optimization
Algorithm
(WOA),
after
sequence
broken
down
into
several
intrinsic
function
(IMF)
using
VMD.
Next,
temporal
each
IMF
captured
a
bidirectional
long
short-term
memory
neural
network
(BiLSTM).
Finally,
all
fused
get
According
experimental
findings,
contrast
most
accurate
baseline
model,
reduced
values
RMSE
18.12,
20.17,
5.36
1∼6
h,
7∼12
13∼24
respectively.
In
addition,
can
successfully
capture
trend
concentration
changes
long-term
prediction.
Atmosphere,
Journal Year:
2025,
Volume and Issue:
16(3), P. 292 - 292
Published: Feb. 28, 2025
PM2.5
in
air
pollution
poses
a
significant
threat
to
public
health
and
the
ecological
environment.
There
is
an
urgent
need
develop
accurate
prediction
models
support
decision-making
reduce
risks.
This
review
comprehensively
explores
progress
of
concentration
prediction,
covering
bibliometric
trends,
time
series
data
characteristics,
deep
learning
applications,
future
development
directions.
article
obtained
on
2327
journal
articles
published
from
2014
2024
WOS
database.
Bibliometric
analysis
shows
that
research
output
growing
rapidly,
with
China
United
States
playing
leading
role,
recent
increasingly
focusing
data-driven
methods
such
as
learning.
Key
sources
include
ground
monitoring,
meteorological
observations,
remote
sensing,
socioeconomic
activity
data.
Deep
(including
CNN,
RNN,
LSTM,
Transformer)
perform
well
capturing
complex
temporal
dependencies.
With
its
self-attention
mechanism
parallel
processing
capabilities,
Transformer
particularly
outstanding
addressing
challenges
long
sequence
modeling.
Despite
these
advances,
integration,
model
interpretability,
computational
cost
remain.
Emerging
technologies
meta-learning,
graph
neural
networks,
multi-scale
modeling
offer
promising
solutions
while
integrating
into
real-world
applications
smart
city
systems
can
enhance
practical
impact.
provides
informative
guide
for
researchers
novices,
providing
understanding
cutting-edge
methods,
systematic
paths.
It
aims
promote
robust
efficient
contribute
global
management
protection
efforts.
Environment Development and Sustainability,
Journal Year:
2024,
Volume and Issue:
unknown
Published: April 2, 2024
Abstract
Air
pollution
is
the
one
of
most
significant
environmental
risks
to
health
worldwide.
An
accurate
assessment
population
exposure
would
require
a
continuous
distribution
measuring
ground-stations,
which
not
feasible.
Therefore,
efforts
are
spent
in
implementing
air-quality
models.
However,
complex
scenario
emerges,
with
spread
many
different
solutions,
and
consequent
struggle
comparison,
evaluation
replication,
hindering
definition
state-of-art.
Accordingly,
aim
this
scoping
review
was
analyze
latest
scientific
research
on
modelling,
focusing
particulate
matter,
identifying
widespread
solutions
trying
compare
them.
The
mainly
focused,
but
limited
to,
machine
learning
applications.
initial
set
940
results
published
2022
were
returned
by
search
engines,
142
resulted
analyzed.
Three
main
modelling
scopes
identified:
correlation
analysis,
interpolation
forecast.
Most
studies
relevant
east
south-east
Asia.
majority
models
multivariate,
including
(besides
ground
stations)
meteorological
information,
satellite
data,
land
use
and/or
topography,
more.
232
algorithms
tested
across
(either
as
single-blocks
or
within
ensemble
architectures),
only
60
more
than
once.
A
performance
comparison
showed
stronger
evidence
towards
Random
Forest
particular
when
included
architectures.
it
must
be
noticed
that
varied
significantly
according
experimental
set-up,
indicating
no
overall
best
solution
can
identified,
case-specific
necessary.
Electronics,
Journal Year:
2024,
Volume and Issue:
13(21), P. 4242 - 4242
Published: Oct. 29, 2024
Accurate
prediction
of
PM2.5
concentration
is
important
for
pollution
control,
public
health,
and
ecological
protection.
However,
due
to
the
nonlinear
nature
data,
accuracy
existing
methods
suffers
performs
poorly
in
both
short-term
long-term
predictions.
In
this
study,
a
deep
learning
hybrid
model
based
on
clustering
quadratic
decomposition
proposed.
The
utilizes
complete
ensemble
empirical
mode
with
adaptive
noise
(CEEMDAN)
decompose
sequences
into
multiple
intrinsic
modal
function
components
(IMFs),
clusters
re-fuses
subsequences
similar
complexity
by
permutation
entropy
(PE)
K-means
clustering.
For
fused
high-frequency
sequences,
secondary
performed
using
whale
optimization
algorithm
(WOA)
optimized
variational
(VMD).
Finally,
temporal
features
are
captured
long-
memory
neural
network
(LSTM).
Experiments
show
that
proposed
exhibits
good
stability
generalization
ability.
It
does
not
only
make
accurate
predictions
short
term,
but
also
captures
trends
prediction.
There
significant
performance
improvement
over
baseline
models.
Further
comparisons
models
outperform
current
state-of-the-art
